Modeling and forecasting from trend-stationary long memory models with applications to climatology

This paper considers the estimation of both univariate and multivariate trend-stationary ARFIMA models, which generate a long memory autocorrelated process around a deterministic time trend. The model is found to be remarkably successful at representing annual temperature and width of tree ring time series data. Forecasts using this model are found to generally be superior to those from AR models. The results indicate an upward trend in temperature that is consistent with global warming; the results for the tree ring series are ambiguous.